Views
No views yet
HF compatible model for Finch-1B6.
! Important Note !The following is the HF transformers implementation of the Finch 1B6 model. This is meant to be used with the huggingface transformers
model = AutoModelForCausalLM.from_pretrained("RWKV/v6-Finch-1B6-HF", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("RWKV/v6-Finch-1B6-HF", trust_remote_code=True)| Eagle 7B | Finch 7B | Finch 14B | |
|---|---|---|---|
| ARC | 39.59% | 41.47% | 46.33% |
| HellaSwag | 53.09% | 55.96% | 57.69% |
| MMLU | 30.86% | 41.70% | 56.05% |
| Truthful QA | 33.03% | 34.82% | 39.27% |
| Winogrande | 67.56% | 71.19% | 74.43% |
1import torch
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4def generate_prompt(instruction, input=""):
5 instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
6 input = input.strip().replace('\r\n','\n').replace('\n\n','\n')
7 if input:
8 return f"""Instruction: {instruction}
9
10Input: {input}
11
12Response:"""
13 else:
14 return f"""User: hi
15
16Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
17
18User: {instruction}
19
20Assistant:"""
21
22
23model = AutoModelForCausalLM.from_pretrained("RWKV/v6-Finch-1B6-HF", trust_remote_code=True).to(torch.float32)
24tokenizer = AutoTokenizer.from_pretrained("RWKV/v6-Finch-1B6-HF", trust_remote_code=True)
25
26text = "请介绍北京的旅游景点"
27prompt = generate_prompt(text)
28
29inputs = tokenizer(prompt, return_tensors="pt")
30output = model.generate(inputs["input_ids"], max_new_tokens=333, do_sample=True, temperature=1.0, top_p=0.3, top_k=0, )
31print(tokenizer.decode(output[0].tolist(), skip_special_tokens=True))1User: hi
2
3Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
4
5User: 请介绍北京的旅游景点
6
7Assistant: 北京是中国的首都，拥有众多的旅游景点，以下是其中一些著名的景点：
81. 故宫：位于北京市中心，是明清两代的皇宫，内有大量的文物和艺术品。
92. 天安门广场：是中国最著名的广场之一，是中国人民政治协商会议的旧址，也是中国人民政治协商会议的中心。
103. 颐和园：是中国古代皇家园林之一，有着悠久的历史和丰富的文化内涵。
114. 长城：是中国古代的一道长城，全长约万里，是中国最著名的旅游景点之一。
125. 北京大学：是中国著名的高等教育机构之一，有着悠久的历史和丰富的文化内涵。
136. 北京动物园：是中国最大的动物园之一，有着丰富的动物资源和丰富的文化内涵。
147. 故宫博物院：是中国最著名的博物馆之一，收藏了大量的文物和艺术品，是中国最重要的文化遗产之一。
158. 天坛：是中国古代皇家1import torch
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4def generate_prompt(instruction, input=""):
5 instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
6 input = input.strip().replace('\r\n','\n').replace('\n\n','\n')
7 if input:
8 return f"""Instruction: {instruction}
9
10Input: {input}
11
12Response:"""
13 else:
14 return f"""User: hi
15
16Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
17
18User: {instruction}
19
20Assistant:"""
21
22
23model = AutoModelForCausalLM.from_pretrained("RWKV/v6-Finch-1B6-HF", trust_remote_code=True, torch_dtype=torch.float16).to(0)
24tokenizer = AutoTokenizer.from_pretrained("RWKV/v6-Finch-1B6-HF", trust_remote_code=True)
25
26text = "介绍一下大熊猫"
27prompt = generate_prompt(text)
28
29inputs = tokenizer(prompt, return_tensors="pt").to(0)
30output = model.generate(inputs["input_ids"], max_new_tokens=128, do_sample=True, temperature=1.0, top_p=0.3, top_k=0, )
31print(tokenizer.decode(output[0].tolist(), skip_special_tokens=True))1User: hi
2
3Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
4
5User: 介绍一下大熊猫
6
7Assistant: 大熊猫是一种中国特有的哺乳动物，也是中国的国宝之一。它们的外貌特征是圆形的黑白相间的身体，有着黑色的毛发和白色的耳朵。大熊猫的食物主要是竹子，它们会在竹林中寻找竹子，并且会将竹子放在竹笼中进行储存。大熊猫的寿命约为20至30年，但由于栖息地的丧失和人类活动的1import torch
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4def generate_prompt(instruction, input=""):
5 instruction = instruction.strip().replace('\r\n', '\n').replace('\n\n', '\n')
6 input = input.strip().replace('\r\n', '\n').replace('\n\n', '\n')
7 if input:
8 return f"""Instruction: {instruction}
9
10Input: {input}
11
12Response:"""
13 else:
14 return f"""User: hi
15
16Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
17
18User: {instruction}
19
20Assistant:"""
21
22model = AutoModelForCausalLM.from_pretrained("RWKV/v6-Finch-1B6-HF", trust_remote_code=True).to(torch.float32)
23tokenizer = AutoTokenizer.from_pretrained("RWKV/v6-Finch-1B6-HF", trust_remote_code=True)
24
25texts = ["请介绍北京的旅游景点", "介绍一下大熊猫", "乌兰察布"]
26prompts = [generate_prompt(text) for text in texts]
27
28inputs = tokenizer(prompts, return_tensors="pt", padding=True)
29outputs = model.generate(inputs["input_ids"], max_new_tokens=128, do_sample=True, temperature=1.0, top_p=0.3, top_k=0, )
30
31for output in outputs:
32 print(tokenizer.decode(output.tolist(), skip_special_tokens=True))
331User: hi
2
3Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
4
5User: 请介绍北京的旅游景点
6
7Assistant: 北京是中国的首都，拥有丰富的旅游资源和历史文化遗产。以下是一些北京的旅游景点：
81. 故宫：位于北京市中心，是明清两代的皇宫，是中国最大的古代宫殿建筑群之一。
92. 天安门广场：位于北京市中心，是中国最著名的城市广场之一，也是中国最大的城市广场。
103. 颐和
11User: hi
12
13Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
14
15User: 介绍一下大熊猫
16
17Assistant: 大熊猫是一种生活在中国中部地区的哺乳动物，也是中国的国宝之一。它们的外貌特征是圆形的黑白相间的身体，有着黑色的毛发和圆圆的眼睛。大熊猫是一种濒危物种，目前只有在野外的几个保护区才能看到它们的身影。大熊猫的食物主要是竹子，它们会在竹子上寻找食物，并且可以通
18User: hi
19
20Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
21
22User: 乌兰察布
23
24Assistant: 乌兰察布是中国新疆维吾尔自治区的一个县级市，位于新疆维吾尔自治区中部，是新疆的第二大城市。乌兰察布市是新疆的第一大城市，也是新疆的重要城市之一。乌兰察布市是新疆的经济中心，也是新疆的重要交通枢纽之一。乌兰察布市的人口约为2.5万人，其中汉族占绝大多数。乌